Data Scoping: Job Search Effort/Behaviour
Ebba Mark
2025-04-28
Overview
The following document summarises current progress on identifying a data source to inform the job search behaviour in the ABM. The following looks into the three determinants of job-finding that we have thus far identified as being relevant (plausibly independent and potentially most relevant):
- Dynamic selection (heterogeneity of job seekers - fitness & skills match for the current labour market)
- Duration dependence (job-finding probability decreases over time spent unemployed indendent of job-seeker characteristics)
- Search effort (individuals can exert greater search effort to counteract duration dependence - generally exhibits a countercyclical behaviour - job search effort increases as the economy recedes)
Current idea: If we view dynamic selection simply as worker heterogeneity, duration dependence as an endogenous process of skill deterioration or signal to employers (up to a certain point), and search effort as responsive to labour market conditions, we will have quasi-independent elements in the job-finding probability of workers in the model!
Goals:
Identify parameters relevant to agent search behaviour in the ABM.
Assess data quality for deriving empirical estimates of these parameters.
Analysis so far:
I first focus on replicating and extending the empirical analysis that speaks to the three determinants above (dynamic selection, duration dependence, search effort)
- (Replicated with additional data for unemployed jobseekers) Mukoyama et al. 2018: Job Search Over the Business Cycle
They provide a novel measure of job search effort exploiting the American Time Use and Current Population Surveys which can be reduced to just the intensive margin (changes in search effort by worker!). At the moment, I think this will be the most useful input for our model.
Abstract: We examine the cyclicality of search effort using time-series, cross-state, and individual variation and find that it is countercyclical. We then set up a search and matching model with endogenous search effort and show that search effort does not amplify labor market fluctuations but rather dampens them. Lastly, we examine the role of search effort in driving recent unemployment dynamics and show that the unemployment rate would have been 0.5 to 1 percentage points higher in the 2008–2014 period had search effort not increased.
(Replicate with additional data from 2017-2024) Eeckhout et al. 2019 Unemployment Cycles
(Replicated with additional data from 2019-2024) Mueller et al. 2021: Job Seekers’ Perceptions and Employment Prospects: Heterogeneity, Duration Dependence and Bias
The authors claim to disentangle the effects of duration dependence and dynamic selection by using job seekers’ elicited beliefs about job-finding. Assuming (and confirming empirically) that job-seekers have realistic initial beliefs about job-finding they isolate the heterogeneity in jobseekers from true duration dependence. Ultimately, they find that dynamic selection selection explains most of the negative duration dependence (rather than pure, true duration dependence).
Findings: Results are remarkably consistent even when including additional data from 2019-2024.
- Survey of Consumer Expectations Reservation Wages, Accepted Wages, and Wage Expectations The data is unfortunately sparse and linking outcomes to reservation wages is difficult. However, in a cross-sectional setting we are able to deduce some weak relationships between Unemployment Duration and Absolute Reservation Wages and Wage Expectations.
Analyses
Eeckhout et al. 2019 Unemployment Cycles
Eeckhout et al. 2019 Unemployment Cycles
## [1] 79
## [1] 0.2000000 0.2315789 0.2631579 0.2947368 0.3263158 0.3578947 0.3894737
## [8] 0.4210526 0.4526316 0.4842105 0.5157895 0.5473684 0.5789474 0.6105263
## [15] 0.6421053 0.6736842 0.7052632 0.7368421 0.7684211 0.8000000
## [1] 10.000000 9.700355 9.324065 8.890497 8.550470 8.226353 7.825428
## [8] 7.521807 7.222388 6.868732 6.645405 6.305140 6.017528 5.710914
## [15] 5.384987 5.117994 4.871024 4.678358 4.598750 4.439575 4.447197
## [22] 4.505201 4.487648 4.486295 4.475880 4.537519 4.619576 4.739504
## [29] 4.801415 4.780796 4.765309 4.688199 4.702700 4.641091 4.628417
## [36] 4.577380 4.472032 4.337180 4.273882 4.208270 4.190000 4.206418
## [43] 4.136236 4.149066 4.129151 4.160927 4.270614 4.443388 4.559865
## [50] 4.825461 5.115954 5.362590 5.571218 5.847741 6.079086 6.244900
## [57] 6.452356 6.597362 6.701002 6.774344 7.012839 7.111058 7.214439
## [64] 7.323705 7.374820 7.483796 7.610651 7.640541 7.735114 7.783887
## [71] 7.786158 7.761714 7.698044 7.608654 7.602465 7.545799 7.547227
## [78] 7.507724 7.382247 7.352305
## [1] 109
## [1] 0.2000000 0.2315789 0.2631579 0.2947368 0.3263158 0.3578947 0.3894737
## [8] 0.4210526 0.4526316 0.4842105 0.5157895 0.5473684 0.5789474 0.6105263
## [15] 0.6421053 0.6736842 0.7052632 0.7368421 0.7684211 0.8000000
## [1] 10.000000 9.700355 9.324065 8.890497 8.550470 8.226353 7.825428
## [8] 7.521807 7.222388 6.868732 6.645405 6.305140 6.017528 5.710914
## [15] 5.384987 5.117994 4.871024 4.678358 4.598750 4.439575 4.447197
## [22] 4.505201 4.487648 4.486295 4.475880 4.537519 4.619576 4.739504
## [29] 4.801415 4.780796 4.765309 4.688199 4.702700 4.641091 4.628417
## [36] 4.577380 4.472032 4.337180 4.273882 4.208270 4.190000 4.206418
## [43] 4.136236 4.149066 4.129151 4.160927 4.270614 4.443388 4.559865
## [50] 4.825461 5.115954 5.362590 5.571218 5.847741 6.079086 6.244900
## [57] 6.452356 6.597362 6.701002 6.774344 7.012839 7.111058 7.214439
## [64] 7.323705 7.374820 7.483796 7.610651 7.640541 7.735114 7.783887
## [71] 7.786158 7.761714 7.698044 7.608654 7.602465 7.545799 7.547227
## [78] 7.507724 7.382247 7.352305 7.288199 7.214835 7.146608 7.098948
## [85] 6.989924 6.946719 6.843443 6.820461 6.631018 6.537070 6.356807
## [92] 6.244352 6.150813 5.455674 7.195513 7.714175 7.912247 7.970527
## [99] 8.006848 7.944453 7.739886 7.468462 7.274557 7.119251 6.971458
## [106] 6.861128 NA NA NA NA
## Linear
## "y~x"
##
## Call:
## lm(formula = as.formula(forms[which(names(forms) == form)]))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.014882 -0.006066 -0.003639 0.007309 0.026123
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.17447 0.03519 -4.958 3.19e-06 ***
## x 0.23294 0.03499 6.656 1.93e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.01036 on 93 degrees of freedom
## Multiple R-squared: 0.3227, Adjusted R-squared: 0.3154
## F-statistic: 44.31 on 1 and 93 DF, p-value: 1.925e-09
##
## Linear with Trend
## "y ~ x + trend"
##
## Call:
## lm(formula = as.formula(forms[which(names(forms) == form)]))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.0102059 -0.0031620 -0.0001317 0.0039334 0.0079867
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -5.268e-02 1.773e-02 -2.970 0.00379 **
## x 1.285e-01 1.731e-02 7.423 5.61e-11 ***
## trend -3.507e-04 1.918e-05 -18.285 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.004839 on 92 degrees of freedom
## Multiple R-squared: 0.8538, Adjusted R-squared: 0.8507
## F-statistic: 268.7 on 2 and 92 DF, p-value: < 2.2e-16
##
## HP Filter
## "y ~ x"
##
## Call:
## lm(formula = as.formula(forms[which(names(forms) == form)]))
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.0068610 -0.0016116 0.0001739 0.0018603 0.0046844
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.049270 0.008149 -6.046 3.05e-08 ***
## x 0.049021 0.008104 6.049 3.02e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.002399 on 93 degrees of freedom
## Multiple R-squared: 0.2824, Adjusted R-squared: 0.2747
## F-statistic: 36.59 on 1 and 93 DF, p-value: 3.018e-08
Mueller et al. Job Seekers’ Perceptions and Employment Prospects: Heterogeneity, Duration Dependence and Bias
Mueller et al: Job Seekers’ Perceptions and Employment Prospects
##
## Descriptive Statistics (SCE)
## ===============================================================
## Variable Orig. 2013-19 2013-24 2020-24
## ---------------------------------------------------------------
## High-School Degree or Less 44.5 40.6 36.9
## Some College Education 32.4 34.9 37.6
## College Degree or More 23.1 24.6 25.6
## Age 20-34 25.4 27.2 30.0
## Age 35-49 33.5 33.6 35.3
## Age 50-65 41.1 39.2 34.8
## Female 59.3 61.2 60.8
## Black 19.1 17.9 16.4
## Hispanic 12.5 13.0 12.6
## UE transition rate 18.7 19.1 18.2
## UE transition rate: ST 25.8 26.5 24.3
## UE transition rate: LT 12.7 12.7 12.3
## # respondents 948 1,367 433
## # respondents w/ at least 2 u obs 534 780 252
## # observations 2,597 3,926 1,347
## ---------------------------------------------------------------
## [1] "Table 2—Regressions of Realized on Elicited 3-Month Job-Finding Probabilities (SCE)"
## [1] "Panel A. Contemporaneous elicitations"
##
## ========================================================================
## Dependent variable:
## ----------------------------------------------------
## T+3 UE Transitions (3-Months)
## Orig. 2013-19 2013-24 2020-24
## (1) (2) (3)
## ------------------------------------------------------------------------
## find_job_3mon 0.464*** 0.396*** 0.265***
## (0.045) (0.036) (0.067)
##
## 1 | userid
##
##
## Constant -0.104 -0.080 -0.136
## (0.169) (0.137) (0.267)
##
## ------------------------------------------------------------------------
## Observations 1,201 1,911 673
## R2 0.218 0.139 0.105
## Adjusted R2 0.207 0.132 0.083
## Residual Std. Error 0.467 (df = 1184) 0.475 (df = 1894) 0.478 (df = 656)
## ========================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
##
## ==========================================================================
## Dependent variable:
## ----------------------------------------------------
## T+3 UE Transitions (3-Months)
## Orig. 2013-19 2013-24 2020-24
## (1) (2) (3)
## --------------------------------------------------------------------------
## find_job_3mon 0.501*** 0.418*** 0.391***
## (0.061) (0.051) (0.094)
##
## findjob_3mon_longterm -0.258*** -0.170** -0.360***
## (0.088) (0.071) (0.133)
##
## longterm_unemployed -0.078 -0.127*** -0.043
## (0.051) (0.041) (0.075)
##
## 1 | userid
##
##
## Constant -0.062 -0.063 -0.402
## (0.175) (0.139) (0.266)
##
## --------------------------------------------------------------------------
## Observations 1,201 1,911 673
## R2 0.259 0.182 0.155
## Adjusted R2 0.248 0.174 0.132
## Residual Std. Error 0.455 (df = 1182) 0.464 (df = 1892) 0.465 (df = 654)
## ==========================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
## [1] "Panel B. Lagged elicitations"
##
## ======================================================================
## Dependent variable:
## --------------------------------------------------
## T+3 UE Transitions (3-Months)
## Orig. 2013-19 2013-24 2020-24
## (1) (2) (3)
## ----------------------------------------------------------------------
## tplus3_percep_3mon 0.332*** 0.241*** 0.203**
## (0.067) (0.056) (0.102)
##
## 1 | userid
##
##
## Constant 0.304 0.490** 0.451
## (0.270) (0.207) (0.394)
##
## ----------------------------------------------------------------------
## Observations 474 798 300
## R2 0.168 0.090 0.179
## Adjusted R2 0.139 0.071 0.132
## Residual Std. Error 0.398 (df = 457) 0.436 (df = 781) 0.447 (df = 283)
## ======================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
##
## ======================================================================
## Dependent variable:
## --------------------------------------------------
## T+3 UE Transitions (3-Months)
## Orig. 2013-19 2013-24 2020-24
## (1) (2) (3)
## ----------------------------------------------------------------------
## find_job_3mon 0.301*** 0.205*** -0.035
## (0.069) (0.058) (0.110)
##
## 1 | userid
##
##
## Constant 0.201 0.422** 0.361
## (0.274) (0.207) (0.400)
##
## ----------------------------------------------------------------------
## Observations 474 798 300
## R2 0.159 0.083 0.168
## Adjusted R2 0.129 0.064 0.121
## Residual Std. Error 0.400 (df = 457) 0.437 (df = 781) 0.450 (df = 283)
## ======================================================================
## Note: *p<0.1; **p<0.05; ***p<0.01
## [1] "Table 4—Linear Regressions of Elicited Job-Finding Probabilities on Duration of Unemployment"
##
##
## Table: Table 4 - Panel A: Linear Regressions of Elicited Job-Finding Probabilities on Duration of Unemployment (SCE)
##
## | | (1) | (2) | (3) | (4) |
## |:-------------------------------|---------:|---------:|---------:|---------:|
## | Unemployment Duration (Months) | -0.0057 | -0.0050 | -0.0043 | 0.0022 |
## | | (0.0007) | (0.0007) | (0.0006) | (0.0049) |
## | Num.Obs. | 882 | 2281 | 2281 | 2281 |
## | R2 | 0.110 | 0.090 | 0.155 | 0.824 |
## | Unemployment Duration (Months) | -0.0050 | -0.0048 | -0.0042 | -0.0026 |
## | | (0.0006) | (0.0006) | (0.0005) | (0.0034) |
## | Num.Obs. | 1265 | 3423 | 3399 | 3423 |
## | R2 | 0.067 | 0.065 | 0.109 | 0.817 |
## | Unemployment Duration (Months) | -0.0011 | -0.0035 | -0.0039 | -0.0077 |
## | | (0.0013) | (0.0012) | (0.0013) | (0.0036) |
## | Num.Obs. | 395 | 1150 | 1140 | 1150 |
## | R2 | 0.002 | 0.019 | 0.118 | 0.838 |
##
## __Note:__
## ^^ Standard errors are clustered at the user or spell level as indicated.
Reservation Wages
Exploring the effect of unemployment duration on reservation wages, accepted wages, and expected wage offers.
Survey of Consumer Expectations Reservation Wages, Accepted Wages, and Wage Expectations (2014-2022) The data is unfortunately sparse and linking outcomes to reservation wages is difficult. However, in a cross-sectional setting we are able to deduce some weak relationships between Unemployment Duration and Absolute Reservation Wages and Wage Expectations.
Draft Plots: Behavioural Parameters Relevant to Working Paper
Finally, I retain the work summarising the:
1.2018/2022 Bureau of Labor Statistics Supplement to the Current Population Survey which asks detailed questions about job search, application effort, and unemployment duration to those who did not opt in to unemployment insurance/compensation. 2. Survey on Consumer Expectations: Which is a “nationallly representative” survey with a Job Search Supplement conducted from 2014-2021.
I provide some preliminary detail on each of these options below including sample size.
Current Population Survey - 2018 Supplement - Summarised
This 2020 “Beyond the Numbers” issue distills insights from a 2018 Supplement to the Current Population Survey. The below plots show the highlights relevant to our decision-making on the job search process. In nearly all cases, the results are “binned” into intervals (ie. number of people sending 81 or more applications or unemployment duration of between 5 and 14 weeks) which means that any line plots (or linear interpretation of the bar graph) should be done carefully. Preliminary results using the raw data are found in the next section.
Figure 1: Shows the proportion of all individuals sending x amount of applications receiving y amount of interviews. The plot indicates a “consistent” return to sending more applications, although as demonstrated in Figure 3, the number of interviews received does not necessarily equate to receiving a job offer.
Figure 2: Demonstrates the number of applications sent (red), interviews received (green), average interview:applicaiton ratio (blue), and probability of receiving a job offer (purple) by individuals in each category of unemployment duration. There is some indication (although, again, interpretation is difficult without the raw data) that both effort and success seems to increase and then decline with time spent in unemployment, apart from success as measured by receiving a job offer which seems to consistently decline with time spent in unemployment.
Figure 3: Percentage of jobseekers receiving an offer seems to increase as a function of the number of applications sent, until a certain point.
Current Population Survey - 2018 & 2022 Supplement - Raw Data
It turns out that the 2018 supplement was also run in 2022, giving us two sets of years to compare (including pre- and post-Covid). The below looks at the raw data that underlies the plotting in the previous section, plus the additional data from 2022. Below find a preliminary scatter plot of applications sent versus unemployment duration. Each individual is asked how many applications they sent in the last two months (two-month periods are indicated by the grey gridlines, for reference). This feels like a promising dataset to me - I think with more careful use of this data this we could approach some more “rigorous” behavioral parameters than simply drawing from the above summary values as calculated by the BLS. This does NOT include data in on the job search. SCE does.
| x |
|---|
| 1 LESS THAN $5,000 |
| 2 5,000 TO 7,499 |
| 3 7,500 TO 9,999 |
| 4 10,000 TO 12,499 |
| 5 12,500 TO 14,999 |
| 6 15,000 TO 19,999 |
| 7 20,000 TO 24,999 |
| 8 25,000 TO 29,999 |
| 9 30,000 TO 34,999 |
| 10 35,000 TO 39,999 |
| 11 40,000 TO 49,999 |
| 12 50,000 TO 59,999 |
| 13 60,000 TO 74,999 |
| 14 75,000 TO 99,999 |
| 15 100,000 TO 149,999 |
| 16 150,000 OR MORE |
Survey on Consumer Expectations - Job Search Supplement
The Federal Reserve Bank of New York compiles the nationally representative Survey on Consumer Expectations annually in October. Since 2013, they have run a Job Search Supplement which includes questions on the time spent searching for work, and unemployment duration. The job search supplement has plenty more questions that we can look at incorporating, listed here. For now, I plot the relationship between time spent searching and time out of work. The table below also indicates the number of people unemployed in the dataset and the number of people unemployed and searching.
| Year | N Unemployed | N Unemp & Searching |
|---|---|---|
| 2014 | 383 | 70 |
| 2015 | 321 | 44 |
| 2016 | 339 | 46 |
| 2017 | 350 | 38 |
| 2018 | 354 | 41 |
| 2019 | 343 | 32 |
| 2020 | 304 | 45 |
| 2021 | 330 | 50 |
On the job search
American Time Use Survey
The American Time Use Survey gives no indication of time spent in unemployment. It shows how much time is spent searching but does not link to time spent in unemployment. Therefore, I prioritised the datasets above. Krueger & Mueller 2010 impute duration spent unemployed from teh ATUS in the following way which could be worth considering.
“Unfortunately, the ATUS interview does not collect information on unemployment duration. Consequently, we derive unemployment duration by taking the unemployment duration reported in the last CPS interview and adding the number of weeks that elapsed between the CPS interview and the ATUS interview. Eighty-six percent of the ATUS interviews were conducted within 3 months of the last CPS interview. For those who were not unemployed at the time of the CPS interview, we impute duration of unemployment by taking half the number of weeks between the CPS and the ATUS interviews. We do not show the weekly LOWESS plot for 13 weeks or less, but simply report the average time allocated to search, as the imputed unemployment duration are quite noisy for those who become unemployed after their last CPS interview.”